Reinforcement Learning & Robotics

In Robotics, RL is one of the most used frameworks for problems such as sequential decision-making and planning under uncertainty. Our research focuses on using sequential decision-making methods, such as Markov Decision Processes (MDPs) and Partially Observable MDPs to plan robot tasks, with an increasing incorporation of model-free learning methods to determine MDP and POMDP policies. Our methods are built to handle systems of increased complexity, with continuous or very large discrete state and action spaces. We are also investigating the use of LLM, VLMs, LRMs and other multimodal language models for task planning. 

Key researchers: Alexandre Bernardino, Francisco Melo, Manuel Lopes, Pedro U. Lima, Plinio Moreno

We explore how robots interact with humans and other robots, often in complex, dynamic, and social environments. 

Social robotics focuses on creating robots that can interact with humans in socially meaningful ways, and that are designed to recognize and respond to human emotions, behaviours, and social cues, making them capable of engaging in natural and effective communication. 

In multi-robot cooperation multiple robots collaborate to perform a task or achieve a common goal. These robots can work together in a coordinated manner, often communicating and sharing information with one another to maximize efficiency and ensure successful task completion. 

Key researchers: Ana Paiva, Pedro Lima, Plinio Moreno